A Q&A with Gabe Newell from Valve, hints on how current 3d cards will perform on HL2 and some slideshow shots from ATI's "Shader Days" event. Expect even more to come in the next days/weeks however in the meantime there is already a bunch of information out of the door for you to check out.

First of all, Valve seems to be extremely proud of their DirectX9 implementation with Half-Life 2 & Microsoft is not unhappy about it either. Now what you wanted to know, although online press benchmarks are not expected until tomorrow, Valve showed off some performance graphs at the aforementioned event... NVIDIA will sure get hurt (this is the kind of bad press no one will ever appreciate) and the fact is, GeForceFX boards are put to shame in HL2 tests, once again Valve assures us the game has a 'pure' DX9 implementation and should represent actual performance of the platform.



That shot (taken from Tech-Report article) shows how boards performed during a normal DX9 test (no special code), Radeon boards totally outperform even NVIDIA's most expensive product.
On a later slide, Valve shows how NVIDIA boards' performance is improved by using a special codepath for the NV3x chips (which they say took a lot of time to build): 5200s and 5600s boards get some ~15% improvement which will probably do little for actual gameplay, while high-end 5900 FX takes a ~70% boost, matching mid-range 9600 Pro performance.

There seems to be little hope for FX boards in next-generation apps, new Dets will most likely help but not enough to change the whole picture... how will the graphics giant manage this? I would say Doom III could be their only hope in order to regain some of gamers' trust, in any case, could two different well-optimized DX9-based titles differ so much (performance-wise) that could affect NVIDIA positively?

Holy cow! Are those cards really that different in architecture?

Thats nuts! Good thing I have a Radeon 9700 Pro!

Well, that makes me less than happy with my week old FX5900U. Guess I can buy a 9800 Pro and give this to my daughter if I have to. I can't think of anything else to say that wouldn't get me banned. I'm stunned!

all i have to say is that cannot be true ....if it is hl2 sucks a## it is too good to be trashed out by atis.

ATi aren't being done any special favours here, as they said. It took them 5 times longer to write this optimized path for NVIDIA cards than it did the default paths. John Carmack has made a few posts where he's said they are using a special NVIDIA optimized path as well.
It's not just Half-Life 2 either, look at pretty much any benchmark out there that utilitzes Pixel Shaders, NVIDIA cards just don't have the performance when it comes to doing them.
NVIDIA can try moan all they want about using the "wrong" Drivers or older HL2 build.
Those new "press" Drivers have issues rendering fog in Half-Life 2 (They can't) & Aquamark 3 as well - namely the image quality is down as compared with earlier drivers. Just another "coincidence" I guess which improved performance. Then of course they go on to say that those benchmarks are invalid as they didn't use those 51.75 drivers (with the bugs) - funny, I don't recall them saying benchmarks were invalid before when using certain drivers, e.g. the ones with the 3d mark 2003 "optimizations" in them, yet now they say their official, whql certified 45.23 Drivers are invalid for testing. Somethings just not right there.